1 / 21

A multilevel approach to geography of innovation

A multilevel approach to geography of innovation. Martin Srholec TIK Centre University of Oslo http://folk.uio.no/martinsr/ DIME International Workshop on

kenyon-hendrix
Download Presentation

A multilevel approach to geography of innovation

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. A multilevel approach to geography of innovation Martin Srholec TIK Centre University of Oslo http://folk.uio.no/martinsr/ DIME International Workshop on "Reconsidering the Regional Knowledge Economy: Theoretical, Empirical and Policy Insights from Diverse Research Approaches", KITE, Newcastle University, 4-5 September 2008, Newcastle, United Kingdom

  2. A multilevel problem • Multilevel modeling concerns hypothesis/problem/data with hierarchical structure • A hierarchy refers to units clustered (grouped/related/similar, etc.) at different levels • Acknowledged in many fields of social science or biology for a long time • School performance is not only given by the amount of study time of a child, but also by higher-level factors such as characteristics of the class, school or national educational system • Similarly the innovation process in firms is influenced by factors operating at different levels

  3. Innovator and environment(s) • Schumpeter focused on constraining factors in the environment: the forces of habit, resistance to innovation by the environment • The idea about survival of firms propelled by innovation, but determined by the environment in the evolutionary economics • National, regional and sectoral innovation systems: organizations, relations and institutions that enable innovation in firms • A need to reach beyond the dichotomy between methodological individualism vs. collectivism in empirical research on innovation • A need for multilevel analysis of interactions between various levels of the analysis in the innovation process

  4. Multilevel modeling • A multilevel model is a statistical model that relates a dependent variable to explanatory variables at more than one level • Sociology, educational research, health science, demography, epidemiology or biology • Not used in research on innovation (and rarely in economics)

  5. Suppose a multilevel model has 2-level structure with firms at level-1 located in regions at level-2. A standard linear 2-level model with one explanatory variable at each level is the following: where yij is the dependent variable, xij is the level-1 predictor, wj is the level-2 predictor, eij, u0j and u1j are random effects (normally distributed residual terms for each equation), i is the firm (i = 1…n) and j is the region (j = 1…m).

  6. Why multilevel modeling? • Single-level models assume that observations (and hence residuals) are independent from each other • Single-level models may suffer from the fallacy of the wrong level • Analyses that use exclusively micro data to account for the contextual effects often suffer from endogeneity • A proper recognition of data hierarchies allows us to examine new lines of questions • And at last but not least the concept of innovation systems implicitly predicts a nested structure of micro data

  7. Why not just dummies? • Using dummies might be a useful quick-fix solution, if we are interested exclusively in the level-1 relationships, but it is of a little help if we are interested in cross-level relations • A dummy is a “catch-all” variable stripped of the context for which we can only speculate what it really represents • If there are many higher-level units, “dummy” models will have many more parameters, decreasing the degrees of freedom • If the higher-level dummies significantly improve predictive power of the model, which indeed is often the case in the literature; a multilevel model should be given priority

  8. Dataset • Micro data from CIS3 (1999-2001) in the Czech Republic • A representative sample of 5,829 enterprises with more than 10 employees was surveyed of which about 65% responded • After omitting firms with incomplete records a dataset of 3,801 firms • Firm-level variables: INNOV, SIZE, AGE and FOREIGN • Factor analysis on 11 regional variables (77 regions at NUTS4)

  9. NUTS4 regions(unemployment rate)

  10. Results of the factor analysis(factor loadings after rotation)

  11. A logit multilevel model Since INNOV is binary, we need to specify a non-linear multilevel model. For this purpose, we distinguish between the sampling model (3.1), a link function (3.2) and a structural part of the multilevel model (3.3). Only the level-1 model differs from the linear case and can be written down as follows: where ij is the log of the odds of success (to introduce an innovation, for instance). Although ij is constrained to be in the interval (0,1), the logit transformation allows ij to take any value and therefore can be substituted to the structural model. Note that the predicted log-odds can be converted to an odds by exp(ij) and to the predicted probability ij by expij/(1+expij).

  12. Basic model Although there are no level-2 predictors, we allow the level-1 intercept and slopes to vary across regions by including the random effects: where there are four fixed effects (00 … 30) and four random effects (u0j … u3j)

  13. Intercept-as-outcome model incorporates the regional variables for RIS and STR into the model as predictors of the level-1 intercept: where we assume that the level-2 predictors explain the different average propensity to innovate across regions, but let the level-1 effects of SIZE, AGE and FOREIGN remain “unconditional” at the regional level.

  14. A typical domestic-owned firm (all of the level-1 effects of zero) located in the best region, which is the Prague agglomeration with the best RIS and modest STR score, is predicted to have 37.3% probability to innovate, whereas the same firm located in the Karviná region with the worst combination of conditions has only 24.4% probability to innovate.

  15. Slopes-as-outcomes model in which we allow the regional variables to influence also slopes of the level-1 predictors: where the new fixed parameters (11 … 31 and 12 … 32) indicate cross-level interactions between the regional and firm-level predictors.

  16. Conclusions • Quality of the regional innovation system directly determines firm’s likelihood to innovate and mediates the effect of some firm-level factors • Also structural problems in the region influence innovation in firms • Just a beginning to show that this approach is viable (another papers in progress) • Main limits given by high demands on the quality (and quantity) of data and computational power of the computer (and the researcher…) • Important implications for policy-making

More Related